��ţӰ��

Basketball coaches and players constantly seek ways to gain an edge over opponents. Imagine a team glancing at a tablet during a timeout, instantly accessing insights about play patterns or player performance, and using that information to shape the next offensive play or defensive formation.

During the summer of 2024 at Duke, the students and faculty of the project are working with the coaching staff of to create a suite of tools that brings real-time analysis of gameplay from basketball video clips.

The team is constructing a computer vision application capable of pinpointing players’ positions on the basketball court. These player coordinates are fed into a machine-learning model. By analyzing player movements, ball passes and court positioning, the tool can identify common offensive and defensive plays.

The information should allow coaches to identify opponents’ play patterns, strengths and weaknesses. During the game, Duke coaches can use the information to see how well the players are executing plays and making strategic adjustments.

Post-game, the data could be useful for assessment of individual player performance, identifying areas for improvement.

The project is ambitious and involves collecting thousands of data points on players from the positioning of their knees and shoulders to how they shift during plays. It is still in the early stages, with team members initially focusing on data extraction of player positioning.

But students are optimistic that the work will result in a tool that aids coaches in scouting opponents and refining play strategies.

The students said the collaborative, interdisciplinary work typical of , part of the , has been a valuable learning experience. But it’s also exciting – and a rare opportunity – to be able to use that learning to maybe provide an advantage for Duke women’s basketball that could make a difference in a close, tense game.

“The idea was a ChatGPT-like system where you could say, ‘here's the information we have about Duke, and here are the plays that they normally run. Can you design a [new] play using the video data?’ Then maybe taking that even a step further like design a specific play for when Duke plays UNC,” said Carlie Scheer, an undergraduate computer science and statistics major.

For more about the project, visit the Rhodes Information Initiative’s